Sparse Datasets Research Articles

Using Deep Learning to Predict Fracture Patterns in Crystalline Solids

Summary Fracture is a catastrophic process whose understanding is critical for evaluating the integrity and sustainability of engineering materials. Here, we present a machine-learning approach to predict fracture processes connecting molecular simulation into a physics-based data-driven multiscale model. Based on atomistic modeling and a novel image-processing approach, we compile a comprehensive training dataset featuring fracture patterns and toughness values for different crystal orientations. Assessments of the predictive power of the machine-learning model shows excellent agreement not only regarding the computed fracture patterns but also the fracture toughness values and is examined for both mode I and mode II loading conditions. We further examine the ability of predicting fracture patterns in bicrystalline materials and material with gradients of microstructural crystal orientation. These results further underscore the excellent predictive power of our model. Potential applications of this model could be widely applied in material design.

Inverse design of broadband highly reflective metasurfaces using neural networks

Metamaterials exhibit optical properties not observed in traditional materials. Such behavior emerges from the interaction of light with precisely engineered subwavelength features built from different constituent materials. Recent research into the design and fabrication of metamaterial-based devices has established a foundation for the next generation of functional materials. Of particular interest is the all-dielectric metasurface, a two-dimensional metamaterial exploiting shape-dependent resonant features while avoiding losses through the use of dielectric building blocks. However, even this simple metamaterial class has a nearly infinite number of possible configurations; researchers now require new methods to efficiently explore these types of design spaces. In this work, we employ rigorous coupled wave analysis to calculate reflection and transmission spectra associated with a class of open-cylinder all-dielectric metasurface. By altering the geometric parameters of open-cylinder metasurfaces, we generate a sparse training data set and construct artificial neural networks capable of relating metasurface geometries to reflection and transmission spectra. Here, we successfully demonstrate that pseudo autodecoder neural networks can suggest device geometries based on a requested optical performance---inverting the design process for this metasurface class. As an example, we query for and discover a particular open-cylinder metasurface displaying a reflection band $R\ensuremath{\ge}99%$ centered at ${\ensuremath{\lambda}}_{0}=1550\phantom{\rule{0.16em}{0ex}}\mathrm{nm}$ that is much broader $\mathrm{\ensuremath{\Delta}}\ensuremath{\lambda}=450\phantom{\rule{0.16em}{0ex}}\mathrm{nm}$ than anything reported for optical metasurfaces. We then analyze the modal interplay in the open-cylinder metasurface to better understand the underlying physics driving the broadband behavior. Ultimately, we conclude that neural networks are ideally suited for generally approaching these types of complex inverse design problems.

Faceted search of heterogeneous geographic information for dynamic map projection

This paper proposes a faceted information exploration model that supports coarse-grained and fine-grained focusing of geographic maps by offering a graphical representation of data attributes within interactive widgets. The proposed approach enables (i) a multi-category projection of long-lasting geographic maps, based on the proposal of efficient facets for data exploration in sparse and noisy datasets, and (ii) an interactive representation of the search context based on widgets that support data visualization, faceted exploration, category-based information hiding and transparency of results at the same time. The integration of our model with a semantic representation of geographical knowledge supports the exploration of information retrieved from heterogeneous data sources, such as Public Open Data and OpenStreetMap.We evaluated our model with users in the OnToMap collaborative Web GIS. The experimental results show that, when working on geographic maps populated with multiple data categories, it outperforms simple category-based map projection and traditional faceted search tools, such as checkboxes, in both user performance and experience.

Evaluation of Four Reanalysis Datasets against Radiosonde over Southwest Asia

Upper-air observational networks in Southwest Asia (SWA) are geographically sparse and reanalysis datasets (RDs) are a typical alternative. However, RDs can perform with varying degrees of quality and accuracy due to differences in assimilation schemes and input observations, among other factors. Geopotential height (gph), air temperature (tmp) and horizontal wind (U and V) modelled by the Japanese 55-year Reanalysis (JRA-55), the European Centre for Medium-Range Weather Forecasts Reanalysis Interim (ERA-I), the ERA fifth-generation (ERA-5), and the National Aeronautics and Space Administration (NASA) Modern Era Retrospective Analysis for Research and Applications version 2 (MERRA), are compared with radiosonde observations at three standard vertical levels (850, 500 and 300 hPa). Results showed that most RDs represent the general climatology, and ERA-5 tended to show the smallest agreements in most cases. RDs did not show consistent performance across seasons, variables, and pressure levels. RDs tended to conduct reasonable estimates over subregions with less complex topography. RDs showed better resampling performance at the upper and lower ends of sounding data distributions more frequently than around the means for most of the variables. This highlights the high potential usefulness of RDs in studying extremes over the region.

Combining instance and feature neighbours for extreme multi-label classification

Extreme multi-label classification problems occur in different applications such as prediction of tags or advertisements. We propose a new algorithm that predicts labels using a linear ensemble of labels from instance- and feature-based nearest neighbours. In the feature-based nearest neighbours method, we precompute a matrix containing the similarities between each feature and label. For the instance-based nearest neighbourhood, we create an algorithm that uses an inverted index to compute cosine similarity on sparse datasets efficiently. We extend this baseline with a new top-k query algorithm that combines term-at-a-time and document-at-a-time traversal with tighter pruning based on a partition of the dataset. On ten real-world datasets, we find that our method outperforms state-of-the-art methods such as multi-label k-nearest neighbours, instance-based logistic regression, binary relevance with support vector machines and FastXml on different evaluation metrics. We also find that our algorithm is orders of magnitude faster than these baseline algorithms on sparse datasets and requires less than 20 ms per instance to predict labels for extreme datasets without the need for expensive hardware.

Inferring effective forces for Langevin dynamics using Gaussian processes.

Effective forces derived from experimental or in silico molecular dynamics time traces are critical in developing reduced and computationally efficient descriptions of otherwise complex dynamical problems. This helps motivate why it is important to develop methods to efficiently learn effective forces from time series data. A number of methods already exist to do this when data are plentiful but otherwise fail for sparse datasets or datasets where some regions of phase space are undersampled. In addition, any method developed to learn effective forces from time series data should be minimally a priori committal as to the shape of the effective force profile, exploit every data point without reducing data quality through any form of binning or pre-processing, and provide full credible intervals (error bars) about the prediction for the entirety of the effective force curve. Here, we propose a generalization of the Gaussian process, a key tool in Bayesian nonparametric inference and machine learning, which meets all of the above criteria in learning effective forces for the first time.

Collaborative filtering recommendation algorithm based on interval-valued fuzzy numbers

Most collaborative filtering recommendation algorithms use crisp ratings to represent the users’ preferences. However, users’ preferences are subjective and changeable, crisp ratings can’t measure the uncertainty of users’ preferences effectively. In order to solve this problem, this paper proposes the interval-valued triangular fuzzy rating model. This model replaces crisp ratings with interval-valued triangular fuzzy numbers on the basis of users’ rating statistics information, which can measure the users’ preferences in a more reasonable way. Based on this model, the collaborative filtering recommendation algorithm based on interval-valued fuzzy numbers is designed. The algorithm calculates the users’ similarity by the interval-valued triangular fuzzy numbers, and takes the ambiguity of ratings into consideration in the prediction stage. Our experiments prove that, compared with other fuzzy and traditional algorithms, our algorithm can increase the prediction precision and rank accuracy effectively with a little time cost, and has an obvious advantage when implemented in a sparse dataset which has more users than items. Thus our method has strong effectiveness and practicability.

Improved covering-based collaborative filtering for new users\u2019 personalized recommendations

User-based collaborative filtering (UBCF) is widely used in recommender systems (RSs) as one of the most successful approaches, but traditional UBCF cannot provide recommendations with satisfactory accuracy and diversity simultaneously. Covering-based collaborative filtering (CBCF) is a useful approach that we have proposed in our previous work, which greatly improves the traditional UBCF and could provide satisfactory recommendations to an active user which often has sufficient rating information. However, different from an active user, a new user in RSs often has special characteristics (e.g., fewer ratings or ratings concentrating on popular items), and the previous CBCF approach cannot provide satisfactory recommendations for a new user. In this paper, aiming to provide personalized recommendations for a new user, through a detailed analysis of the characteristics of new users, we reconstruct a decision class to improve the previous CBCF and utilize the covering reduction algorithm in covering-based rough sets to remove redundant candidate neighbors for a new user. Furthermore, unlike the previous CBCF, our improved CBCF could provide personalized recommendations without needing special additional information. Experimental results suggest that for the sparse datasets that often occur in real RSs, the improved CBCF significantly outperforms those of existing work and can provide personalized recommendations for a new user with satisfactory accuracy and diversity simultaneously.

Recommending Curated Content Using Implicit Feedback

Matrix factorization (MF) which is a Collaborative filtering (CF) based model, is widely used in the recommendation systems (RS). For our experiment, we collected data from a company's internal web site where curated contents are published and pushed to the employees. However, the size of the dataset is small and interaction data is also limited. We got a sparse matrix when we generated a user-item rating matrix. We have used Multi-Layer Perceptron (MLP) to calculate the rating scores from the implicit feedbacks. However, on this sparse dataset traditional content only or CF-only RSs do not work well. Here, we propose ahybrid RS that incorporates content similarity scores into an MLP-based MF-model. To integrate the content similarity scores into the MF, we have defined an objective function based on a regularization term. The experimental result shows that our proposed model demonstrates a better result than the traditional MF-based models.

Classification with costly features as a sequential decision-making problem

This work focuses on a specific classification problem, where the information about a sample is not readily available, but has to be acquired for a cost, and there is a per-sample budget. Inspired by real-world use-cases, we analyze average and hard variations of a directly specified budget. We postulate the problem in its explicit formulation and then convert it into an equivalent MDP, that can be solved with deep reinforcement learning. Also, we evaluate a real-world inspired setting with sparse training dataset with missing features. The presented method performs robustly well in all settings across several distinct datasets, outperforming other prior-art algorithms. The method is flexible, as showcased with all mentioned modifications and can be improved with any domain independent advancement in RL.

Multivariate Modelling of the Trace Element Chemistry of Arsenopyrite from Gold Deposits Using Higher-Dimensional Algebras

In geochemistry, the elevated concentrations of certain elements in rock or mineral samples are used for the assessment of a mineralising system’s important geological characteristics. The relationship between the elements and system characteristics may take one of two forms, regression or classification. The collection and analysis of geochemical samples typically leads to the development of sparse and fuzzy multivariate datasets, which often have complex interrelationships. As a result, it is difficult to quantitatively and qualitatively interpret many of these geochemical datasets and relate the resulting data to relevant geological characteristics. There are no governing expressions in the current literature that can relate mineral-based trace element assemblages to geological characteristics. In this work, a generalised formulation for multivariate systems modelling is presented and it is applied to the discovery of the manifested system characteristics of arsenopyrite samples collected from Australian gold (Au) deposits, using fuzzy arsenopyrite trace element datasets. Arsenopyrite trace element assemblages, collected using laser ablation-inductively coupled plasma-mass spectrometry, are represented as spatially distributed networks embedded in $$ {\mathbf{R}}^{4,1} $$ conformal geometric algebraic space. The interactions of elements within these networks generate multivectors which are referred to as hyperfields. The functions and parameters used to generate these hyperfields are optimised to create multiple unique models, which are used to discover relationships between system characteristics and the multivariate dataset. The models generated allow a specialist to discover qualitative relations, using higher-dimensional representations of multivariate data. In this work, the generalised hyperfield formulation is applied to the problem of predicting the endowment, lithology and camp of Au deposits, using an arsenopyrite trace element dataset. The hyperfield formulation is compared against the state-of-the-art random forest decision tree algorithm ‘XGBoost’ and artificial neural networks (ANNs).

The Tall Tower Dataset: a unique initiative to boost wind energy research

Abstract. A dataset containing quality-controlled wind observations from 222 tall towers has been created. Wind speed and wind direction measurements covering the 1984–2017 period have been collected from existing tall towers around the world in an effort to boost the utilization of these non-standard atmospheric datasets, especially within the wind energy and research fields. Observations taken at several heights greater than 10 m above ground level have been retrieved from various sparse datasets and compiled in a unique collection with a common format, access, documentation and quality control. For the last, a total of 18 quality control checks have been considered to ensure the high quality of the wind records. Non-quality-controlled temperature, relative humidity and barometric pressure data from the towers have also been obtained and included in the dataset. The Tall Tower Dataset (Ramon and Lledó, 2019a) is published in the repository EUDAT and made available at https://doi.org/10.23728/b2share.136ecdeee31a45a7906a773095656ddb.

Abstract OT1-03-02: Combining MRI and liquid biopsies for prediction of response to neoadjuvant chemotherapy in early stage breast cancer: The LIMA study

Abstract Background The response to neoadjuvant chemotherapy (NAC) in early stage breast cancer has important prognostic implications1-3. Early, dynamic prediction of response allows for adaption of the treatment plan before completion, or even before the start of treatment. This strategy can help prevent overtreatment and related toxicity and correct for undertreatment with an ineffective regimen. We hypothesize that accurate dynamic response prediction may be reached by combining multi-parametric MRI with liquid biopsies prior to, during and after NAC, in addition to conventional clinical and pathological information. Magnetic resonance imaging (MRI) is non-invasive and is typically used for response evaluation in current clinical practice. It shows the size and perfusion of the tumor as they change during treatment. However, tumor size on MRI has limited predictive value for response to therapy4. Multi-parametric MRI uses different imaging protocols in one session to measure more functional items than perfusion alone, addressing different aspects of tumor biology, and possibly improving predictive value. With this improvement, imaging still only visualizes macroscopic disease. Therefore, in the LIMA study, MRI will be combined with liquid biopsies containing circulating tumor cells (CTCs) and circulating tumor DNA (ctDNA), which have both shown prognostic and predictive values in early stage breast cancer5-7. Since the ctDNA may originate from cells in every part of the tumor, it may capture tumor heterogeneity. Liquid biopsies are minimally invasive and provide insight into microscopic tumor load and the tumor’s genetic picture. There is an abundance of clinical studies investigating the predictive value of liquid biopsies and MRI. On breast cancer however, no studies have been published that combine these two modalities. Aim show proof of concept for combining multi-parametric MRI with liquid biopsies in addition to conventional clinical and pathologic information, to accurately predict response to NAC at different time points. Trial design A multicenter prospective observational cohort study. Multi-parametric MRI will be performed prior to NAC, halfway and after completion of NAC. Liquid biopsies will be obtained before start of treatment, every 2 weeks during treatment and after completion of NAC. Statistical methods The primary endpoint is Residual Cancer Burden (RCB) in the surgical resection specimen after NAC. Collected data will be analyzed with efficient statistical methods dedicated to sparse datasets primarily using penalized regression techniques. Eligibility criteria Non-pregnant, non-lactating women ≥18 years of age with non-metastatic, histologically proven invasive breast cancer who are planned to be treated with NAC (and in case of a HER2-positive tumor: addition of trastuzumab and/or pertuzumab) who have: - no luminal A breast cancer - no inflammatory breast cancer - no other active malignant disease in the past 5 years (excluding squamous cell or basal cell carcinoma of the skin) -no contra-indications for MRI or gadolinium-based contrast agents Target accrual 100 patients (present: 0) References 1von Minckwitz 2012 J Clin Oncol; 2Cortazar 2014 Lancet; 3Symmans 2007 J Clin Oncol; 4Gu 2017 Clin Breast Cancer; 5Bidard 2018 J Natl Cancer Inst; 6Garcia-Murillas 2015 Sci Transl Med 7Magbanua 2018 SABCS p. Abstract No. PD2-01 Acknowledgements Funding from the European Union Horizon 2020 research and innovation program under grant agreement no. 755333 (LIMA). In collaboration with Philips, INSERM, ICM, Agena Bioscience, ANGLE, Stilla Technologies, DiaDx and ALS Automated Lab Solutions. Citation Format: Liselore Maria Janssen, Britt Berendine Maria Suelmann, Markus Hendrik Adriaan Janse, Sjoerd Geert Elias, Paul J van Diest, Wouter Bernard Veldhuis, Elsken van der Wall, Kenneth George Antonius Gilhuijs. Combining MRI and liquid biopsies for prediction of response to neoadjuvant chemotherapy in early stage breast cancer: The LIMA study [abstract]. In: Proceedings of the 2019 San Antonio Breast Cancer Symposium; 2019 Dec 10-14; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2020;80(4 Suppl):Abstract nr OT1-03-02.

I-Eclat: performance enhancement of eclat via incremental approach in frequent itemset mining

One example of the state-of-the-art vertical rule mining technique is called equivalence class transformation (Eclat) algorithm. Neither horizontal nor vertical data format, both are still suffering from the huge memory consumption. In response to the promising results of mining in a higher volume of data from a vertical format, and taking consideration of dynamic transaction of data in a database, the research proposes a performance enhancement of Eclat algorithm that relies on incremental approach called an Incremental-Eclat (i-Eclat) algorithm. Motivated from the fast intersection in Eclat, this algorithm of performance enhancement adopts via my structured query language (MySQL) database management system (DBMS) as its platform. It serves as the association rule mining database engine in testing benchmark frequent itemset mining (FIMI) datasets from online repository. The MySQL DBMS is chosen in order to reduce the preprocessing stages of datasets. The experimental results indicate that the proposed algorithm outperforms the traditional Eclat with 17% both in chess and T10I4D100K, 69% in mushroom, 5% and 8% in pumsb_star and retail datasets. Thus, among five (5) dense and sparse datasets, the average performance of i-Eclat is concluded to be 23% better than Eclat.

A Recommendation Method for Highly Sparse Dataset Based on Teaching Recommendation Factorization Machines

There is no reasonable scientific basis for selecting the excellent teachers of the school’s courses. To solve the practical problem, we firstly give a series of normalization models for defining the key attributes of teachers’ professional foundation, course difficulty coefficient, and comprehensive evaluation of teaching. Then, we define a partial weight function to calculate the key attributes, and obtain the partial recommendation values. Next, we construct a highly sparse Teaching Recommendation Factorization Machines (TRFMs) model, which takes the 5-tuples relation including teacher, course, teachers’ professional foundation, course difficulty, teaching evaluation as the feature vector, and take partial recommendation value as the recommendation label. Finally, we design a novel Top-N excellent teacher recommendation algorithm based on TRFMs by course classification on the highly sparse dataset. Experimental results show that the proposed TRFMs and recommendation algorithm can accurately realize the recommendation of excellent teachers on a highly sparse historical teaching dataset. The recommendation accuracy is superior to that of the three-dimensional tensor decomposition model algorithm which also solves sparse datasets. The proposed method can be used as a new recommendation method applied to the teaching arrangements in all kinds of schools, which can effectively improve the teaching quality.

Improving collaborative filtering's rating prediction coverage in sparse datasets by exploiting the 'friend of a friend' concept

In collaborative filtering users with highly similar tastes are termed 'near neighbours' and recommendations for a user are based on her 'near neighbours' ratings. However, for a number of users no near neighbours can be found, a problem termed as the 'grey sheep' problem. This problem is more intense in sparse datasets. In this work, we propose an algorithm for alleviating this problem by exploiting the friend of a friend (FOAF) concept. The proposed algorithm, CFfoaf, has been evaluated against eight widely used sparse datasets and under two widely used collaborative filtering correlation metrics, namely the Pearson Correlation Coefficient and the Cosine Similarity and has been proven to be particularly effective.

Improving collaborative filtering's rating prediction coverage in sparse datasets by exploiting the 'friend of a friend' concept

Improving collaborative filtering's rating prediction coverage in sparse datasets by exploiting the 'friend of a friend' concept

D-GENE: Deferring the GENEration of Power Sets for Discovering Frequent Itemsets in Sparse Big Data

Sparseness is the distinctive aspect of big data generated by numerous applications at present. Furthermore, several similar records exist in real-world sparse datasets. Based on Iterative Trimmed Transaction Lattice (ITTL), the recently proposed TRICE algorithm learns frequent itemsets efficiently from sparse datasets. TRICE stores alike transactions once, and eliminates the infrequent part of each distinct transaction afterward. However, removing the infrequent part of two or more distinct transactions may result in similar trimmed transactions. TRICE repeatedly generates ITTLs of similar trimmed transactions that induce redundant computations and eventually, affects the runtime efficiency. This paper presents D-GENE, a technique that optimizes TRICE by introducing a deferred ITTL generation mechanism. D-GENE suspends the process of ITTL generation till the completion of transaction pruning phase. The deferral strategy enables D-GENE to generate ITTLs of similar trimmed transactions once. Experimental results show that by avoiding the redundant computations, D-GENE gets better runtime efficiency. D-GENE beats TRICE, FP-growth, and optimized versions of SaM and RElim algorithms comprehensively, especially when the difference between distinct transactions and distinct trimmed transactions is high.

Bootstrapping a Dense Distribution from a Sparse Distribution – a Bitcoin Returns Experiment

By modifying a standard bootstrap resampling technique, we can take sparse daily return data sets and generate much more dense data sets that are plausibly similar to the original data in distribution shape and outlier probability. Examples are provided for Bitcoin daily returns that have large outliers.

Nonconvex Sparse Representation With Slowly Vanishing Gradient Regularizers

Sparse representation has been widely used over the past decade in computer vision and signal processing to model a wide range of natural phenomena. For computational convenience and robustness against noises, the optimization problem for sparse representation is often relaxed using convex or nonconvex surrogates instead of using the l 0 -norm, the ideal sparsity penalty function. In this paper, we pose the following question for nonconvex sparsity-promoting surrogates: What is a good sparsity surrogate for general nonconvex systems? As an answer to this question, we suggest that the difficulty of handling the l 0 -norm does not only come from the nonconvexity but also from its gradient being zero or not well-defined. Accordingly, we propose desirable criteria to be a good nonconvex surrogate and suggest a corresponding family of surrogates. The proposed family of surrogates allows a simple regularizer, which enables efficient computation. The proposed surrogate embraces the benefits of both l 0 and l 1 -norms, and most importantly, its gradient vanishes slowly, which allows stable optimization. We apply the proposed surrogate to wellknown sparse representation problems and benchmark datasets to demonstrate its robustness and efficiency.